04/Nov/03 Liliana Martin PROGRESS 10 X 1K collections loaded of the 'PR' and 'SP" . The loading time much more faster (less "Smkdir" to create a collection name). 03/Nov/03 Liliana Martin PLAN Tests with the new collection names. Less "Smkdir" commands used to create a collection name. Query 1 and 100 fixed collections to see what the response is every 1000 collections of the type 'PR' then another of the type 'SP' 30/Oct/03 Liliana Martin PROGRESS The 310000 collections loaded. Tests stopped. 23/Oct/03 Liliana Martin PROGRESS The loading is always slow . It has taken more then 10 hours to load 5K for 2 days. I decide to query 1 collection 5 times and a fixed set of collections 1 time after each 5K loaded. 21/Oct/03 Liliana Martin SOLUTION|IDEA It has seemed that the script "load2srb" is longer and longer since 16 Oct 2003. Now it takes always more than 8 hours , before it took less than 6 hours if no problem. The load script with 12 x 5K collections has started on Friday and it is always running. How many have srbServer been running on bbrmcat1 for 5 days ? I don't know. For now here are many ones connected to : bbr-test-xfer1.slac.stanford.edu, bigsur17.roma1.infn.it, babar.fzk.de:50434,objyserv01.slac.stanford.edu But perhaps it is due to the quering which was changed ? See explications below We had seen that to query 1 collection had been longer (only 1 iterate) than to query a bunch of them(only 1 iterate too). So on 16 Oct 2003 I started to query : a) 1 collection, always the same (skim199/0009/00099900/14.0.0a/skim199_00099999_V01_V02) ; the query is repeated 10 times after each 5K loaded b) a fixed set of collections ( 100 collections, always the same); the query is repeated 3 times after each 5K loaded. About a) it seems that the first repeat is almost always higher then the others ones. the range is 4-54 sec ; The others repeats take 3-4 sec. It's strange . About b) no significative differences beteewn the iterates(3) after each 5K; by the way, the number of second slightly increase with more 5K 08/Oct/03 Adil Hasan PROGRESS Liliana is currently running her tests and timing how long it takes to query collections. The format of her collection queries is: 07/Oct/03 Adil Hasan PLAN Created the project file. The purpose of this project is to find out how the MCAT scales with increasing numbers of collections. The initial part of the test is to understand the scaling with collections with an arbitrary number of levels. The second part of the test is to test scaling with collections of a much lower number of levels. 03/Oct/03 Liliana PROGRESS 02/Oct/03 Liliana PROGRESS The loading program is now running on flora05.slac.stanford.edu. The load is much more faster ( ~20000 sec). 01/Oct/03 Liliana PROGRESS The loading program is running on ccasn02.in2p3.fr . The load of the first 5K is very slow (73404 sec).